jrutil VS fquery

I've also tried to rewrite some of my challenging queries [0] in my hypothetical syntax and while I think your observation about pipeline length is correct, the result still came out much better than SQL. Frankly, even in F#, most of my pipelines are around 5 functions too. In my view, pipelines are just a convenient mental model. I'd love to see your sketches, here are my (very WIP) concepts: [1]

[0]: for example this monstrosity: https://gitlab.com/dvdkon/jrutil/-/blob/dbb971c18526e68dcc97...

Similar benefits without codegen (based on decorator magic) for a python based stack:

https://github.com/adsharma/fquery

* Use dataclasses for both database schema and the user facing operations

Re: Conways law at Facebook

I was at Facebook when GraphQL was invented, maintaining a backend storage service where a core assumption was that storage should be reorganized based on access patterns and that predicates should be pushed down to storage where they can be executed more efficiently.

GraphQL was hard to push predicates down, because you don't know which of the edges were written in PHP.

My response was fquery[1], which is like what's being discussed here but with python as the source language instead of swift and amenable to preserving the largest possible query structure for backend optimizers, including SQL optimizers.

It has some early demos converting a GraphQL/fquery into SQL where possible. It should be possible to add enough metadata to fquery to identify if an edge is non-trivial (calls into another microservice) or trivial (can be optimized to a storage backend or SQL).

[1] https://github.com/adsharma/fquery